fix(driver): shlex-split --cflags so spaced defines survive

[MarkAtwood]

capture_macros tokenised CFLAGS with str.split(), so a define with a quoted spaced value (-DBANNER="a b") was torn into -DBANNER="a + b" and the tail dropped — silently corrupting the build-configuration evidence recorded in the SBOM. Now uses shlex.split (failing loudly on an unbalanced quote) so the define stays a single token.

Regression test added to test_sbom.py (guarded on a host compiler): asserts -DWG_BANNER="a b c" reaches the compiler intact as #define WG_BANNER a b c. Independent of the other open review PRs.

Addresses SBOM-gpex.8.

[Copilot]

Pull request overview

This PR fixes SBOM build-evidence capture by correctly tokenizing --cflags using shell-style parsing so quoted, space-containing -D macro values (e.g., -DWG_BANNER="a b c") survive as a single compiler argument instead of being silently corrupted.

Changes:

• Switch capture_macros() in share/sbom-driver.py from str.split() to shlex.split(), and fail fast on malformed quoting.
• Add a regression unit test ensuring a spaced -D value reaches the compiler intact (skips cleanly if no host C compiler is available).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
tests/test_sbom.py	Adds a host-compiler-gated regression test verifying spaced -D macro values are preserved through capture_macros().
share/sbom-driver.py	Uses shlex.split() for --cflags parsing and exits with a clear error on unbalanced quotes to avoid silent SBOM evidence corruption.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[MarkAtwood]

Closing. shlex is a net regression here, and the splitter is the wrong layer to fix this at.

Measured against the actual split_cflags inputs:

input master .split() this PR (shlex)
-DMSG="hello world" BROKEN (2 tokens) correct
-I"/opt/my sdk/inc" BROKEN (2 tokens) correct
-DPICO_SDK_PATH=C:\Users\ci\sdk correct BROKEN -> C:Userscisdk
-I\\server\share\inc correct BROKEN -> \servershareinc
-D"__WOLFBOOT" (wolfBoot today) works works

Neither wins. Each fixes one class of input by breaking the other, and shlex.split(..., posix=False) is not a middle ground: it only groups when a quote starts the token, so for -DMSG="hello world" it behaves identically to plain .split().

The real defect is upstream of the splitter. CFLAGS is already a token list in both Make and CMake; flattening it into one string and re-lexing it is what introduces the ambiguity. We already solved exactly this for source lists with --srcs-file.

Fix: add --cflags-file, one token per line, written directly by the build fragment. No quoting, no lexing, no host-shell or host-OS dependence, and it strengthens the reproducibility guarantee rather than trading one breakage for another. The --cflags string form stays for compatibility on .split() semantics and gets deprecated.

Not urgent. The only spaced define in wolfBoot today is -D"__WOLFBOOT", an even quote count with no internal space, so .split() handles every configuration we currently build.

Follow-up: master's split_cflags docstring argues correctly against shlex and then implies plain split is correct. It should name the spaced-value limitation and point at --cflags-file.